Add gl global min #1
Conversation
| @@ -4,7 +4,7 @@ include("parameter.jl") | |||
| function record_current(io, i, j, estimate, estimates) | |||
| if io != nothing | |||
| prop = j/((i-1)*ne+j) | |||
| est = (1-prop)*estimates + (prop)*mean(estimates[1:j]) | |||
| est = (1-prop)*estimate + (prop)*mean(estimates[1:j]) | |||
There was a problem hiding this comment.
this might be the bug we are looking for @Chamusssss !
| OuterLevelTwisting(N, C, S, μ, innerlevel, Ψ, initialguess, n_restarts) | ||
| #initialguess = zeros(S) | ||
| #n_restarts = 20 | ||
| n_restarts = 2 |
| end | ||
|
|
||
|
|
||
| get_result(t::OuterLevelTwisting) = t.μ | ||
| set_result!(t::OuterLevelTwisting, μ) = (t.μ[:] = μ) | ||
|
|
||
|
|
||
| function twist!(t::OuterLevelTwisting, parameter::Parameter) | ||
| function twist_local!(t::OuterLevelTwisting, parameter::Parameter) |
There was a problem hiding this comment.
try merge twist_local and twist_global into one twist!. Specifically, might be good to add a flag in struct OuterLevelTwist that multiplex the two options. And in twist! implement different behavior based on that flag
| @@ -261,7 +312,7 @@ end | |||
| ⋅ Z ∼ N(μ, I) where μ is shifted mean that minimizes variance | |||
| ⋅ W ∼ q where q is shifted bernoulli probability that minimizes upper bound on the variance | |||
| """ | |||
| function glassermanli_mc(parameter::Parameter, sample_size::Tuple{Int64, Int64}, extra_params=(nothing, nothing), io::Union{IO, Nothing}=nothing) | |||
| function glassermanli_mc_local(parameter::Parameter, sample_size::Tuple{Int64, Int64}, filename:: String, extra_params=(nothing, nothing)) | |||
There was a problem hiding this comment.
not a good idea to use filename as a mandatory argument to the function. What if I just want to see the result without saving intermediates into that file? The reason for the io argument previously was so that if you desire to see the intermediates, then supply that as an optional parameter to the function and omit it otherwise.
| @@ -261,7 +312,7 @@ end | |||
| ⋅ Z ∼ N(μ, I) where μ is shifted mean that minimizes variance | |||
| ⋅ W ∼ q where q is shifted bernoulli probability that minimizes upper bound on the variance | |||
| """ | |||
| function glassermanli_mc(parameter::Parameter, sample_size::Tuple{Int64, Int64}, extra_params=(nothing, nothing), io::Union{IO, Nothing}=nothing) | |||
| function glassermanli_mc_local(parameter::Parameter, sample_size::Tuple{Int64, Int64}, filename:: String, extra_params=(nothing, nothing)) | |||
There was a problem hiding this comment.
Would definitely want to merge local and global together.
- if we have decided on which one to use, then hard-code the default as is
- if we decide that we will do experiment comparing the two methods, then would supply a flag via
extra_paramsthat decides on if to uselocalorglobal. i.e.extra_param=(<value_for_mu> | "local" | "global", <value_for_sigma> | nothing). Or something like this
No description provided.